8
“Fingerprint classification using relational graph” Index: 1>Introduction 2> Preprocessing 3>Relational graph construction 4> Graph based classification 5>Conclusion 1. Introduction :- Automated fingerprint identification has been the most widely accepted bio- metric authentication method for last decades. Role of the fingerprint clas- sification in any fingerprint identification system is of crucial importance because it reduces the processing time taken in identification of

Fingerprint Classification

Embed Size (px)

Citation preview

Page 1: Fingerprint Classification

“Fingerprint classification using relational graph”

Index:

1>Introduction

2> Preprocessing

3>Relational graph construction

4> Graph based classification

5>Conclusion

1. Introduction:-Automated fingerprint identification has been the most widely accepted bio-

metric authentication method for last decades. Role of the fingerprint clas-

sification in any fingerprint identification system is of crucial importance

because it reduces the processing time taken in identification of fingerprint,

especially when the number of fingerprints registered in the system is large.

Classification allows an input fingerprint to be matched against only by a

subset of a database and is critical in not only speeding-up the rest of the

identification process but increasing the matching accuracy. However, per-

formance of fingerprint classification can be easily affected by the nature of

Page 2: Fingerprint Classification

fingerprint image distortions such as varying appearance and quality by ro-

tation and skin condition. Furthermore, extracting informative features is

getting harder as fingerprint scanning devices are getting smaller.

A structural classification algorithm is proposed by intro-

ducing the novel definition of a fingerprint relational graph. Graph-based

techniques have been known as powerful and convenient tools to deal with

the above mentioned problem, but sensitive to deformation of fingerprints .

2. Preprocessing:-For construction of a relational graph, all informative features are collected

from an input image through following preprocessing steps: computing the

orientation field, specifying the core area, and segmenting the orientation

field.

2.1 Orientation field computation:-Computation of the orientation field is performed . An input finger-

print image is divided into a number of small blocks of the same size and the

dominant orientation is estimated in each block. The size of each block is

determined based on the resolution of acquired fingerprint images. (In our

implementation, the block size was set to 8 × 8.) The area containing no

ridge in the input image is eliminated in this step.

2.2 Segmentation of orientation field:-The segmentation step is for partitioning the directional image according

to the orientation fields computed. In the process, a connected component

Page 3: Fingerprint Classification

analysis scheme is applied, that is to determine regions containing blocks with

the same directions. In the implementation, the orientation field is quantized

into 4 directions (0 ◦ , 45 ◦ , 90 ◦ , 135 ◦ ) beforehand. Then, for an initial block

selected, neighbor blocks with the same directions are grouped and labeled.

The process is repeated until every block in the orientation field is labeled.

Once the first scan is completed, small regions created due to spurious minu-

tiae are merged into the most adjacent one.

2.3 Core area specification:-The core area defined in our implementation is located where the distribution

of the ridge directions is expected to be more reliable. Therefore, consistent

location of the core area for all types of fingerprints is required. The center

point, detected by the Jain et al.s algorithm , is used as the registration

point to specify the core area. A part of the orientation field around the

center point is zoned for the core area, with an arbitrary initial size. The size

of the area is iteratively reduced so that none of the background is included.

When the final size of the area is smaller than a certain predefined threshold,

the fingerprint under examination is rejected as the clues collected from the

area are not sufficient.

3. Relational graph construction:-The construction of relational graphs from the segmented image is performed

by the following steps. If R = {R 1 , R 2 , ..., R n } is the set of regions of the segmented image, we

Page 4: Fingerprint Classification

define the relational graph of R as G = (V, E, f V C , f V d , f E )where,

• V = {v i |i = 1, ..., n} is a set of nodes, each corresponds to R i .

• E = {e ij |1 ≤ i < n, i < j ≤ n, Adjacency(R i , R j )} is a set of undirected

edges; Adjacency(R i , R j ) returns binary value depending if R i and R j

are adjoining.

• f V c (v i ) represents the overlapped R i with the core area. When f V c (v i )=0,

which means that any part of R i does not covered by the core area, and

R i is much smaller than other regions, v i and related edges are elimi-

nated .

• f V d (v i ) represents a relative direction determined by Eq. (1). f V d (v i ) = |θ (v i ) − θ (v max ) |.

4.Graph-based classification:-For the classification, a type of association, between an input graph and one

or more predefined model graphs built from training set, is required. As

distances among graphs cannot be obtained directly in analytic closed-forms,

a lot of attempts for inexact graph matching have been proposed. One of

the most flexible means to express the graph similarity is the edit distance.

The basic idea of the graph edit distance is to find the minimum number of

transformations required from a graph to another. In our implementation, a

fast algorithm proposed by Reisen and Bunke is employed for computing

Page 5: Fingerprint Classification

the edit distance. In the mean time, a set of model graphs, each of which

represents the graph structure of each class properly, has to be built from

the training set.

5.Conclusion:-We presented the graph-based classification algorithm with the novel defi-

nition of the fingerprint relation graph and the edit distance computation

for the cost function. Since the newly defined graphs are constructed by

taking account into clues in the core area actively, there is no need to an

algorithm-specific guideline for users to control image quality in acquisition

with various fingerprint impressions. Concentrating on the most reliable area

is appropriate to recent small touch scanners. The experimental results show

that the practical approach proposed performs better than other structural

fingerprint classification approaches.

Page 6: Fingerprint Classification

***************************************